Comparing Imacon X5, Creo iQ3 and Sony A7R IV scans

[albireo]

To finally throw the dedicated scanner color superiority myth out the window once and for all.

Besides, this thread is about resolution. The images aren't color-balanced to match, they're Negmaster defaults. I posted several disclaimers above.

It's not a myth, and it's not a scanner vs camera challenge. It's an inherent difference between line sensors and camera sensors, backed by solid theory. Great engineering behind both technologies, though designed for slightly different purposes. I'd be all over a factory-tuned camera rig employing a non-interpolating sensor. Somebody in this thread mentioned a D800E, sounds like a good starting point for 35mm scanning at least.

I'm personally not, in general, against camera scanning. In fact I do see 2 obvious advantages over dedicated film scanners: speed, once the setup is in place and provided pixel shifting and stitching are not of interest, and silence. My Nikon Coolscan makes an awful racket. I'd love to replace it with something quiet. But I just like its results and the associated easy workflow too much.

So I won't replace it with a digital camera and all the surrounding paraphernalia. I still remember my shock when I first saw the out-of-camera jpegs from my Fujifilm XT-2, back when I was still using digital cameras. The Xtrans array configuration was so difficult to deconvolve that even the OOC jpegs showed heavy artefacts (widely known as 'painterly effect' or 'worms'). I'm sure Fujifilm has addressed these issues with the later generation of cameras/software, but just to be sure an X-trans sensor is not something I'd want near my negatives.

You mention resolution as opposed to colour rendition as the target of your test. Genuine question and not trying to be flippant: I just wonder why, if pursuit of ultimate resolution is your goal, why employ a hybrid film setup at all? Wouldn't your Sony Alpha be capable of ultimate, easy to obtain resolution figures if used as a camera and not as a film scanning device? Unless of course your goal is to maximise resolution of historical film material for scanning purposes (where I can see a case for this type of setup).

 

[GLS]

Never tried ColorPerfect, but I use both NLP and Negmaster. Their authors have very different philosophies towards color inversion, if you will.

1. NLP takes the aggressive guessing approach: give me an image, no matter what, and I will try to guess what the most pleasing colors for it should be. It is optimized for one-click simplicity and speed. In the latest version you feed it an entire roll, click a single button, and you'll get mostly decent results by default. I am not surprised it's so popular.
2. Negmaster takes "I am RA4 paper" approach, for the lack of a better description. Its algorithm feels far simpler, but more consistent. It is extremely sensitive to digitization exposure. So much so that I recommend exposure bracketing. Its workflow is not as quick. First step is to apply a custom DCP profile which strips Adobe color science [1], then you trasnfer the file into Photoshop and run the plugin. Batch processing is so primitive that it's fair to say it's absent.

Both of them give you a 16-bit image to apply further tweaks to, but Negmaster output is more malleable and colors don't jump all over from scene to scene like they do with NLP.

Usually I approach them this way:

• Apply the Negmaster DCP profile in Lightroom, open in Photoshop and invert manually. That's the golden standard. No automatic tool can beat that. That's how I easily matched color from all of these scanners. For high quality "portfolio" images this is where it stops. Everything below is to speed up batch processing for vacation rolls or everyday snapshots.
• Select a sequence of shots of the same scene / light conditions and run them through NLP. If the output is close to the reference image above, I am done.
• If some images aren't quite there, I run them through Negmaster conversion and then tweak them into shape in PS, usually just a couple of slight curves.

There's also Negmaster BR, which is a completely different product. I have not tried it because it's meant to be used with scanners only. Knowing the author, I'm sure it applies a similar color philosophy.

If I were to choose one, it would be Negmaster.

P.S. I am also fond of grain2pixel. It produces really nice flat & desaturated TIFFs that are easy to edit. It is also free. My only criticism of it is that it applies additional tweaks to the image, completely removes color noise, for example. It also makes all film emulsions look more or less the same. Its batch mode is pretty good too.

[1] I believe that Adobe color fuckery is the reason for people complaining about cameras not capable of capturing true film colors. If you strip it away, you get the same uninverted image as what X5 or Creo would give you.

Thanks for this.

Yes, I have noticed the results from NLP can vary quite a lot. Batch processing isn't a concern for me as I don't shoot 35mm, so don't have dozens of images to trawl through. If I get 5 or 6 keepers from a roll of 120 I'm doing well.

The DCP workflow from Negmaster is what has me most interested tbh, as I know their potential (I made my own custom DCP for camera scanning of chromes, and the results are night and day better than the Adobe defaults).

 

[Steven Lee]

It's not a myth, and it's not a scanner vs camera challenge. It's an inherent difference between line sensors and camera sensors, backed by solid theory. Great engineering behind both technologies, though designed for slightly different purposes. I'd be all over a factory-tuned camera rig employing a non-interpolating sensor. Somebody in this thread mentioned a D800E, sounds like a good starting point for 35mm scanning at least.

Let me disagree with you here. But before I elaborate further, let me emphasize that I have zero feelings towards one technology or another, I am just a genuinely curious person who enjoys tinkering and who started his career in digital data acquisition and signal processing. So I happen do have a decent degree of understanding of how all of this works.

Now... I am not sure what do you mean by inherent difference between line sensors and camera sensors, backed by solid theory. The underlying sensor tech is the same, with a solid 10+ years of additional R&D in favor of modern camera sensors. If you are referring to different light frequencies filtered out by Bayer vs linear, I couldn't find any specs online, but we can probably analyze what they are and adjust in processing. Perhaps you are referring to the demosaicing artifacts? Just enable the 4-step pixel shift mode that removes those, but TBH I find them negligible.

Basically I am not aware of any inherent problems backed by solid theory that would prevent a modern camera sensor and processing algorithms from picking up the entire gamut of CMY densities present on film and repackaging them into an sRGB color space. Instead, I see inherently identical technologies, and yet people online spread the notion of how it's supposedly impossible to get the same color out of a digital mirrorless camera. Well... my patience ran out one day, so I went ahead and did exactly that: picked the most expensive dedicated film scanners and got the same color out of my camera

You mention resolution as opposed to colour rendition as the target of your test. Genuine question and not trying to be flippant: I just wonder why, if pursuit of ultimate resolution is your goal, why employ a hybrid film setup at all? Wouldn't your Sony Alpha be capable of ultimate, easy to obtain resolution figures if used as a camera and not as a film scanning device? Unless of course your goal is to maximise resolution of historical film material for scanning purposes (where I can see a case for this type of setup).

That is a great question! Thank you for asking. I do not care about resolution at all. I love shooting film precisely because it's imperfect, and I do not print big. There aren't enough walls in my small house I posted these samples because some folks asked me for them earlier. I do like nice colors though. But nice is not the same as accurate. Film is perfect for that, and I wouldn't want to leave them on the table when I'm scanning, so I was mainly evaluating color in these tests. There's one aspect of high-resolution scanning that matters sometimes. I am talking about the grain character for certain types of B&W images. But that's another digression.

 

[albireo]

Now... I am not sure what do you mean by inherent difference between line sensors and camera sensors, backed by solid theory. The underlying sensor tech is the same, with a solid 10+ years of additional R&D in favor of modern camera sensors. If you are referring to different light frequencies filtered out by Bayer vs linear, I couldn't find any specs online, but we can probably analyze what they are and adjust in processing. Perhaps you are referring to the demosaicing artifacts? Just enable the 4-step pixel shift mode that removes those, but TBH I find them negligible.

So then, if you find those differences negligible, that's another matter altogether. Ultimately what matters is what YOU see and what YOU expect.

I was only commenting as I had thought you were trying to offer your results with a sprinkling of scientific rigour and in an attempt to provide a pretty general result, in which case whether you find the differences negligible or not would be irrelevant, because the differences are there. Let me focus on the last one you mentioned, the demosaicising artifacts.

Off on a tangent - as you I'm sure know already, the raw output of Bayer-filter cameras consists of a so-called Bayer pattern image: an arrangement of colour filters on a square grid of photosensors. In the Bayer arrangement this filter consists of a matrix of repeating 2x2 pixel patterns, one coding for red, one for blue, and two for green. Importantly, each pixel is filtered to record only one of three colors:

The key thing here is that each pixel of the sensor is behind a colour filter and the output is an array of pixel values, each indicating a raw intensity for one of Red, Green or Blue. This arrangement needs an algorithm to estimate for each pixel the colour levels for all colour components, rather than a single component.

This is called `demosaicising'. There are different implementations of this - it is in essence a flavour of signal interpolation. Now compared to the initial, raw intensity images, the reconstructed image is typically accurate in uniform-coloured areas, but will have a significant loss of spatial accuracy and many would agree colour accuracy in complex regions.

But to go back to scanning, dedicated film scanners do not rely on Bayer (or worse, X-trans) pattern matrices and the raw output they produce does not require demosaicing. The so-called `line CCD sensors' in a scanner are, at a very raw level, better than any camera sensor because they do not interpolate and because they use only a single line using the best part of a sharp dedicated lens, so there is no optical distortion or other lens flaws added.

One consequence of the lack of a Bayer array+demosaicing is that when a scanner like the Coolscan 8000/9000 is scanning 90mp, those are 90mp of full color data. Digital camera color data is only 1/4 of the stated resolution due to the above. So even, say, a Fujifilm GFX 100 (a 102mp sensor, 8K$ camera) is only getting 25mp of full color data (and another 25mp of extra green [luminosity] data) from its 100mp of photosites.

You mention a workaround to limit the above: pixel shift. It's a great way to improve your results, but it comes with issues, some of which have already been mentioned above. And you are still left with the limitations of digital camera colour, any lens flaws, having go through the hassle of stitching when scanning 120 or above, plus any other issues inherent with the specific home-made scanning setup used (vibrations of the repro stand? imprecise sensor/film alignment? poor quality/evenness of the retroillumination; and much more). Orange mask removal is another story and so is the lack of IR (infra-red channel) for dust removal in home made DSLR scanning rigs.

Why the wall of text? It's really not aimed directly at you @Steven Lee . I just keep stumbling, on the broader social media, on content trying to sell DSLR scanning as THE thing to do if one wants to really enjoy hybrid photography. There is I think considerable commercial interest in selling overpriced scanning kit, hip 3D printed $500 holders, etc. There are famous bloggers out there with direct commercial interest in DSLR scanning gadgets (you know who they are) and youtube vloggers paid by DSLR scanning companies to 'upgrade' to DSLR scanning. The results is that many beginners, teenagers who just purchased an AE-1 program and want to jump into film are starting to believe that a $4000 DSLR-based setup is what it's going to give them those awesome, professional results they crave. Hint: it is not. Exposure and processing understanding is far more important imo for the film experience (even when film is digitalised, as scanned film can fully preserve some of the characteristics many of us love about film). When exposure and development is fine tuned, a $200 Plustek or a refurbished Coolscan used correctly will provide considerable enjoyment to many people our there without breaking the bank. I really like thinking that film photography is a hobby for everyone who can afford to buy film, a camera, a Paterson tank, and little more

Over and out and thanks for the interesting test!

 

[Helge]

all. I love shooting film precisely because it's imperfect, and I do not print big. There aren't enough

Now see, this is the kind of quasi religious notions I don’t get coming out of a man who clearly knows better.
Film is not inherently especially imperfect. A CMOS sensor is absolutely as imperfect as film, it’s an analog device (like most sensors) that has some of the same limitations of film and some completely different ones.

Resolution is important. And for many reasons too. “Not caring about resolution” is too general and sweeping a statement to be useful.

If there is some more general, meta takeaway from this thread, it’s that the different methods really have quite different outputs.
There are artifacts in those scans that can’t be attributed to different profiles or noise reduction.
And that neither the scanners or the film have “bottomed out” the film.

Scanning of any current type, has a way of making you think that they extracted all there was. That goes for crap like the Scanza, the Epsons and the Imacons.
It’s the deceptive artifacting and pseudo authoritative, neatly packaged output.
But still dye clouds are not even marginally distinguishable, and the different levels of detail extracted here, points to there being more.

So then, if you find those differences negligible, that's another matter altogether. Ultimately what matters is what YOU see and what YOU expect.

I was only commenting as I had thought you were trying to offer your results with a sprinkling of scientific rigour and in an attempt to provide a pretty general result, in which case whether you find the differences negligible or not would be irrelevant, because the differences are there. Let me focus on the last one you mentioned, the demosaicising artifacts.

Off on a tangent - as you I'm sure know already, the raw output of Bayer-filter cameras consists of a so-called Bayer pattern image: an arrangement of colour filters on a square grid of photosensors. In the Bayer arrangement this filter consists of a matrix of repeating 2x2 pixel patterns, one coding for red, one for blue, and two for green. Importantly, each pixel is filtered to record only one of three colors:

The key thing here is that each pixel of the sensor is behind a colour filter and the output is an array of pixel values, each indicating a raw intensity for one of Red, Green or Blue. This arrangement needs an algorithm to estimate for each pixel the colour levels for all colour components, rather than a single component.

This is called `demosaicising'. There are different implementations of this - it is in essence a flavour of signal interpolation. Now compared to the initial, raw intensity images, the reconstructed image is typically accurate in uniform-coloured areas, but will have a significant loss of spatial accuracy and many would agree colour accuracy in complex regions.

But to go back to scanning, dedicated film scanners do not rely on Bayer (or worse, X-trans) pattern matrices and the raw output they produce does not require demosaicing. The so-called `line CCD sensors' in a scanner are, at a very raw level, better than any camera sensor because they do not interpolate and because they use only a single line using the best part of a sharp dedicated lens, so there is no optical distortion or other lens flaws added.

One consequence of the lack of a Bayer array+demosaicing is that when a scanner like the Coolscan 8000/9000 is scanning 90mp, those are 90mp of full color data. Digital camera color data is only 1/4 of the stated resolution due to the above. So even, say, a Fujifilm GFX 100 (a 102mp sensor, 8K$ camera) is only getting 25mp of full color data (and another 25mp of extra green [luminosity] data) from its 100mp of photosites.

You mention a workaround to limit the above: pixel shift. It's a great way to improve your results, but it comes with issues, some of which have already been mentioned above. And you are still left with the limitations of digital camera colour, any lens flaws, having go through the hassle of stitching when scanning 120 or above, plus any other issues inherent with the specific home-made scanning setup used (vibrations of the repro stand? imprecise sensor/film alignment? poor quality/evenness of the retroillumination; and much more). Orange mask removal is another story and so is the lack of IR (infra-red channel) for dust removal in home made DSLR scanning rigs.

Why the wall of text? It's really not aimed directly at you @Steven Lee . I just keep stumbling, on the broader social media, on content trying to sell DSLR scanning as THE thing to do if one wants to really enjoy hybrid photography. There is I think considerable commercial interest in selling overpriced scanning kit, hip 3D printed $500 holders, etc. There are famous bloggers out there with direct commercial interest in DSLR scanning gadgets (you know who they are) and youtube vloggers paid by DSLR scanning companies to 'upgrade' to DSLR scanning. The results is that many beginners, teenagers who just purchased an AE-1 program and want to jump into film are starting to believe that a $4000 DSLR-based setup is what it's going to give them those awesome, professional results they crave. Hint: it is not. Exposure and processing understanding is far more important imo for the film experience (even when film is digitalised, as scanned film can fully preserve some of the characteristics many of us love about film). When exposure and development is fine tuned, a $200 Plustek or a refurbished Coolscan used correctly will provide considerable enjoyment to many people our there without breaking the bank. I really like thinking that film photography is a hobby for everyone who can afford to buy film, a camera, a Paterson tank, and little more

Over and out and thanks for the interesting test!

While your sentiment and thinking is fine and applaudable it somewhat misses some of the points of the thread.

- It would be very easy to build a very good scanner, based on camera scanning ideas with all the superfluous stuff and even harming stuff (like Bayer filter) out, with cheap components made readily available by the smartphone industry, for someone like Plustek.
In other words we (me?) are trying to gain more general insight into the fundamental problems.

- It’s perfectly possible that line sensors are a fundamentally better idea in theory.
They are however not made on new processes and with new technology. Perhaps they are even taken from an old depleting production run made years ago?
And they are slow in use. In the time a sweep takes, you can take twenty matrix sensors shots, merge them and get even better results.

You forgot an important problem with Bayer sensors:
The dyes in the array are not fully saturated RGB. That would bring sensitivity down and most importantly make demosaicing even harder.
That’s why it would be wonderful with a monochrome sensor and spectrally peaky backlight. Both for black and white and colour.

 

[Steven Lee]

@Helge I get what you're saying, but I just fail to feel excited about fine detail. Look, a solid 80% of my images have a bit of motion or shake blur or missed focus in them. On top of that, my vision is not what it used to be, so subjectively I am just not drawn to the crispness you're referring to, although I feel it when I see it. As I said earlier, my 24MP Fuji delivered more detail than I ever needed, and I used my 36MP DSLR on the "medium RAW" setting because large files are a PITA to move around via WiFi, and large JPEG files sometimes take a couple of seconds to render full detail in some album apps, including Apple Photos.

Speaking about imperfect film's interpretation of the world, we're looking at highlights compression in CN films, B&W grain texture, etc. I like those. We all have weird preferences, heh?

 

[Helge]

@Helge I get what you're saying, but I just fail to feel excited about fine detail. Look, a solid 80% of my images have a bit of motion or shake blur or missed focus in them. On top of that, my vision is not what it used to be, so subjectively I am just not drawn to the crispness you're referring to, although I feel it when I see it. As I said earlier, my 24MP Fuji delivered more detail than I ever needed, and I used my 36MP DSLR on the "medium RAW" setting because large files are a PITA to move around via WiFi, and large JPEG files sometimes take a couple of seconds to render full detail in some album apps, including Apple Photos.

Speaking about imperfect film's interpretation of the world, we're looking at highlights compression in CN films, B&W grain texture, etc. I like those. We all have weird preferences, heh?

Ok. Fair enough. And again, thank you for cropping and posting the scans. I always find it fascinating to see other peoples results. And it’s rare to be able to compare these scanners.

 

[Adrian Bacon]

[1] I believe that Adobe color fuckery is the reason for people complaining about cameras not capable of capturing true film colors. If you strip it away, you get the same uninverted image as what X5 or Creo would give you.

I know I'm super late to this discussion, and tend to have my own biases, but this is one of the primary reasons I moved my own scanning code away from generating DNG files natively to just generating flat TIFF files. Adobe has what appears to be some pretty hard coded baked in things it does to the images that you have to jump through hoops to either get around or undo, and personally, it was just easier to drop doing the Adobe thing all together and just use libraw to parse the raw camera files, and libtiff to output plain vanilla 16 bit TIFF files.

 

[Steven Lee]

@Adrian Bacon

 

[Lachlan Young]

I'm surprised that we've got this far and no-one has considered the obvious and simple comparative test: an inherently non-bayer grayscale only sensor in a mirrorless or DSLR (whether original or converted), with a suitable set of RGB filters swung in place for each sequential exposure (add an IR if you want to remake the wheel with ICE incorporated too) - after all, there's a fair few scanners from back in the not-so-distant past that did this for speed of output (Kodak RF3570 for example), but were limited in outright resolution by the sensors of the early-mid 90's.

And while we're on this, the sensor in a Frontier scanner is a shift & stitch. None of this is new, and the problems were solved (with compromises that we now have the ability to resolve) in the recent past.

 

[brbo]

I'm surprised that we've got this far and no-one has considered the obvious and simple comparative test: an inherently non-bayer grayscale only sensor in a mirrorless or DSLR (whether original or converted), with a suitable set of RGB filters swung in place for each sequential exposure (add an IR if you want to remake the wheel with ICE incorporated too) - after all, there's a fair few scanners from back in the not-so-distant past that did this for speed of output (Kodak RF3570 for example), but were limited in outright resolution by the sensors of the early-mid 90's.

I've considered it many years ago, but I don't have 2k EUR lying around to buy the cheapest BW-sensor camera with still acceptable resolution.

I have done the best approximation of such process. And also a silly approximation of a BW sensor where I've used BW film instead. I've shown it here on Photrio...

And while we're on this, the sensor in a Frontier scanner is a shift & stitch. None of this is new, and the problems were solved (with compromises that we now have the ability to resolve) in the recent past.

Frontiers do not do shift & stitch, they do pixel-shift. Some hi-end flatbeds do shift & stitch.

 

[Lachlan Young]

I've considered it many years ago, but I don't have 2k EUR lying around to buy the cheapest BW-sensor camera with still acceptable resolution.

Which is less money than most of the good scanners + supplies, incipient associated Apple museums etc would cost - and if we reckon that 24-ish mp single-shot seems plenty for most, that opens up a lot of options, at least for getting a functional proof-of-concept system - as much as a 33x44mm sensor would be good to have down the line. And something like the Intrepid enlarger light source would probably deliver more than good enough RGB without needing to fiddle with filters etc. If the camera could spit out a direct linear .tiff output like the Kodak DCS760m etc managed over 20 years ago, that would also make things much easier...

Frontiers do not do shift & stitch, they do pixel-shift. Some hi-end flatbeds do shift & stitch.

Yeah, I corrected my post - brain remembered the pixel-shift mechanism, fingers typed shift & stitch... Either way, current pixel shift is nothing new.

 

[brbo]

Which is less money than most of the good scanners + supplies, incipient associated Apple museums etc would cost - and if we reckon that 24-ish mp single-shot seems plenty for most, that opens up a lot of options, at least for getting a functional proof-of-concept system...

Not everything people do gets posted hero on Photrio. "Trichromatic" scans (+ IR channel) with true BW cameras or cameras with sensors striped of CFA, IR filters have been around for a while now. I guess the market is still to small to entice a mass production of a device that would employ an array sensor instead of still much cheaper line sensor.

 

[Lachlan Young]

Not everything people do gets posted hero on Photrio. "Trichromatic" scans (+ IR channel) with true BW cameras or cameras with sensors striped of CFA, IR filters have been around for a while now. I guess the market is still to small to entice a mass production of a device that would employ an array sensor instead of still much cheaper line sensor.

If you're referring to versions of the Phase One Cultural Heritage system and the like, yes they've been around for a while (not that Phase makes much of the hardware themselves - a lot of it is Cambo) - and the way Phase's market has shifted almost completely towards certain industrial/ institutional/ governmental sectors is very telling about who's actually buying their stuff these days. I also think the bigger problem is that it's not really a resolution issue, but more a colour reproduction and MTFsystem one for the results that most people need/ want - contrast reproduction and colour matter much more than outright pixel count.

 

[brbo]

No, I'm referring to enthusiasts (I don't think Phase One system is doing it right (monochrome sensor)) with home made rigs for capturing 3 or 4 (IR) separate captures on monochrome sensor.

How good are the results? Great in regard of colour reproduction. Getting insane resolution is a bit harder because you need highly corrected lens, perfect alignment, zero vibration, stitching, etc... Basically, it's reinventing a proper scanner, a few decades later.

 

[Carnie Bob]

If you're referring to versions of the Phase One Cultural Heritage system and the like, yes they've been around for a while (not that Phase makes much of the hardware themselves - a lot of it is Cambo) - and the way Phase's market has shifted almost completely towards certain industrial/ institutional/ governmental sectors is very telling about who's actually buying their stuff these days. I also think the bigger problem is that it's not really a resolution issue, but more a colour reproduction and MTFsystem one for the results that most people need/ want - contrast reproduction and colour matter much more than outright pixel count.

I have the Phase One Cultural system here in my lab , I also owned an Imocan and do have an Eversmart Supreme scanner.... I have worked and am working on all three systems daily, even a Epson V850 system.. our business is printing in historical process from collections from various sources.

I believe there are subtle differences in every system one uses and quality expectations of each... I. am definitely not a print sniffer and I actually find that aspect of what we do quite amusing..

The new phase system is very good, edge to edge I think it is spectacular, as is the Creo. for smaller formats I would prefer the Phase.
The Imocan was by far the easiest system to work with and was a go to for many years in my work. Larger format films would go on the Creo

Now I am convinced that the Phase equals or betters most systems I have used, There are drawback for sure but not something an old dog cannot learn. I have been
so reliant on Imocan and Creo slow antiquated software that moving to Phase has been an eyeopener and learning curve for me.

I am not getting to be confident with Capture One and it looks like a great software system that is not like PS that I used heavily with the other two older systems.

As Lachian mentions this system is aimed towards people who can afford the system (govt) so there is not a lot of them in private hands like myself. I would say
its a solid system. better in some ways than other but I still like working with an Epson for quick scans. for simple website work.